Fundamentals of Machine Learning
Ce cours (IFT6390) étant conçu pour les étudiants au 2e et 3e cycles, la langue parlée par la majorité des étudiants est l'anglais. Le cours sera donc enseigné dans cette langue.
Il bien sûr possible de poser des questions en français, tout comme de répondre aux examens et devoirs dans la langue de Molière. Une version pour le 1er cycle (enseignée en français) est offerte également
par Guillaume Rabusseau. Pour la version française de la cette page, consultez cette page.
This course targets graduate students. Hence, I will teach in English for greater accessibility. You can of course ask questions in French, and hand in your assignments and exams in French. An undergrad version of this course is also offered in French by Guillaume Rabusseau in the fall.
Virtual Tools
This course is taught live on Google Meet. Recordings and written notes are then uploaded to a Google Drive folder. I only use Studium for uploading grades. For live chat, join the dedicated Slack workspace.
Schedule
- Monday: 12:30 to 2:30PM
- Wednesday: 2:30PM to 3:30PM
Topics
- Recap: linear algebra, calculus
- General terminology
- Supervised learning
- Classification and regression problems
- Unsupervised learning
- Density estimation
- K-neareset neighbors
- Dimensionality reduction
- Statistical learning principles
- Capacity, generalization and model complexity
- Overfitting
- Cross-validation
- K-folds cross-validation
- Double descent phenomenon
- Optimization principles
- Unconstrained optimization
- Necessary conditions for optimality
- Necessary and sufficient conditions
- Constrained optimization
- Necessary conditions for optimality
- Necessary and sufficient conditions
- Algorithms
- Newton's method
- Unconstrained optimization: steepest descent
- Derivation as a constrained optimization problem
- Optimization with constraints
- Sequential Quadratic Programming
- Statistical estimators
- Maximum likelihood estimator
- Maximum a posteriori (MAP)
- Classical linear models
- Linear regression
- Logistic regression
- Linear SVM
- Classical non-linear models
- Artificial neural networks
- Automatic differentiation
- Forward mode
- Reverse mode
- Recurrent neural network
- Decision trees, random forests
- Unsupervised learning
- Meta-learning
Wellbeing